Antlr Tree Grammar to Generate Pmachine code?

Question

I am presently working on a project to build a small c compiler to run on a pmachine. Presently, i started off using a stripped version of a full c grammar. Parsing looks okay at the moment and i used the rewrite rules to build the AST which looks kinda okay. Presently, i am in the phase of building a tree grammar to walk and generate pmachinecode. My question is,

Will the tree grammar be a feasible choice to walk the tree and generate p-code ? (e.g i run an expression and return a string of pcode) or is the approach of building the AST manually without Antlr the right way to achieve this ?

I just want to know if i am really on track using the tree grammar or not.

Your advice will be much appreciated.

Best regards.

Ira Baxter Ira Baxter · Accepted Answer · 2013-04-29T20:38:03

If your intention is to generate stack machine P code, walking the AST will probably do fine. Don't expect the resulting code to be stellar.

You can generate code that is more focused on (abstract or real) registers with an AST walk, and lot of peephole optimizations. (Basically you generate abstract code but delay emitting it as long as possible; the peephole optimizer looks at sequences of abstract code and finds useful optimations and emits those instead.

Example: abstract code emitted:

  PUSH X
  PUSH Y
  ADD

Peephole optimizer recognizes these and produces (additional Pmachine instructions) LOAD ACC, X ADD ACC, Y PUSH ACC

This avoids 3 memory reads and writes. If you pmachine has a very slow interpreter, you might not care, but as it gets faster, this overhead starts to be significant.

You can easily invent dozens of useful peephole optimizations.

Antlr Tree Grammar to Generate Pmachine code?

1 Answers